TemplaitExamples
What a template looks like when it is written down
Three templates, each as the list of zones it is made of: what each part is, what it is for, and the instruction that goes with it. The first one is close to the template a new account starts from.
A square launch post - 1080 x 1080
Four zones, a plate behind the headline, a photograph and a small table. This is the shape a template takes when the design is simple: almost all of the work is in the sentences, not in the arrangement.
| Zone | Kind | Role | Instruction |
|---|---|---|---|
| Hero block | Shape | Decoration | A flat plate behind the headline. No content. |
| Headline | Text | Hero | One sentence, under twelve words, present tense, no exclamation marks. |
| Product visual | Image | Product showcase | A photograph of the product on a plain background, lit from the left. |
| Benefits grid | Table | Evidence | Four rows: the benefit, then one clause saying what it does for the reader. |
An event poster - A3, 300 DPI
A print preset, so the export is sized for paper rather than for a screen. The zones are the ones a poster has to answer: what it is, when and where, why to come, and how to get in.
| Zone | Kind | Role | Instruction |
|---|---|---|---|
| Event title | Text | Hero | The event name, as it should read on a wall. |
| Date and venue | Text | Informational | Date first, then the venue on the second line. No comma between them. |
| Key visual | Image | Decoration | A wide photograph, dark enough for white type to sit on it. |
| What happens | List | Informational | Five items, each under eight words. |
| Registration | QR code | Call to action | Encodes the ticket page address. |
| Organised by | Logo | Brand identity | The organiser mark, with the name beside it. |
Two of these zones carry no text the assistant can write - the QR code encodes an address you give it, and the logo is the organiser's own mark. Declaring a role for them anyway is what stops an assistant from inventing words to fill them.
An infographic - A2, 300 DPI
Here the roles do the heavy lifting. Two metrics, a chart and two paragraphs are all “content”, and the only thing distinguishing the headline figure from the source line is what each zone is declared to be for.
| Zone | Kind | Role | Instruction |
|---|---|---|---|
| Title | Text | Hero | A question, phrased so the numbers below answer it. |
| Headline figure | Metric | Evidence | One percentage, rounded, with the change since last year. |
| Second figure | Metric | Evidence | One count in thousands, spelled out rather than abbreviated. |
| Trend chart | Chart | Evidence | Bars, six periods, axes labelled in the units the figures above use. |
| Explanation | Text | Informational | Two sentences saying what the numbers mean. No italics. |
| Source | Text | Legal disclaimer | The dataset name, its year, and the address it came from. |
The same post, as the export you paste
A template is designed in the app and used everywhere else, so the artifact that leaves is a document. This is the launch post above, exported as Markdown and trimmed to the two parts that matter: the guide in front, and the elements.
# Task Generate exactly one completed visual artifact matching the specified canvas type, dimensions, and aspect ratio. Read the entire specification before generating. … # Global Instructions - Interpretation: Treat the canvas, variables, attachments, and elements as one cohesive specification. Resolve variable references and use relevant attachments as instructed. - Composition: Render all elements together in one artifact. Preserve element count, layout, layering, and intentional empty space. … … # Launch Banner Brief <!-- The elements above and below are the layout, back to front. … --> ## Elements ### 1. Hero block - Type: shape - Semantic role: decoration ### 2. Headline - Type: text - Semantic role: hero <!-- The brief this element is written from: the instruction, then the limits. --> #### Generation - Instruction: One sentence, under twelve words, present tense, no exclamation marks. ### 3. Product visual - Type: image - Semantic role: product-showcase #### Generation - Instruction: A photograph of the product on a plain background, lit from the left. ### 4. Benefits grid - Type: table - Semantic role: evidence #### Generation - Instruction: Four rows: the benefit, then one clause saying what it does for the reader.
The whole thing is pasted as it is. The guide at the top is addressed to the assistant and tells it to treat the document as one composition - removing it is the most common way to get four unrelated answers instead of one design. The comments are notes to whoever is reading, and they say nothing the document does not already say.
What the three have in common
- One zone, one job. A zone asking for two things gets one of them, and you will not be able to tell which until you look.
- Every zone has an instruction, even the ones carrying no words. An undeclared zone is a box, and an assistant fills a box with whatever it likes.
- Roles are declared rather than implied by size or position, so the emphasis in the design is stated rather than guessed at from geometry.
- Limits are written down - a word ceiling, a row count, a rounding rule. They are the difference between a paragraph and the paragraph you needed.
Build one
Start from a blank canvas or from the starter template, and change the words rather than the structure to begin with - the instructions are what most templates need work on.